Been spending a lot of time in Polymarket's short-duration BTC/ETH markets (the 5-min and 15-min ones) and noticed something interesting: the order book doesn't always keep up with the underlying price.

Here's the mechanism: these markets resolve based on where BTC/ETH is trading on major exchanges (like Binance) at expiry. But Polymarket's own order book is driven by human/bot market participants placing limit orders — and in the final seconds before resolution, there's often a lag between where the book is priced and where the actual spot price has moved to.

That gap is the entire game in these markets. It closes fast (often single-digit seconds), which is why manual trading struggles here — by the time you've read the price and clicked, the window's usually gone.

This is exactly the kind of microstructure problem that's fun to study, whether or not you ever automate it:

• Watch the spread between Binance spot and the implied Polymarket price as expiry approaches

• Notice how order book depth thins out in the last 10-15 seconds

• Track how often the "lagging" side actually gets filled vs. reprices before you can act

I ended up building a small educational toolkit around this exact problem (watches Binance in real time, flags the lag, and handles execution with Fill-Or-Kill orders so you're never left holding a stale fill) — mostly because I wanted to actually test the paper-trading numbers instead of just theorizing. Happy to answer questions on the mechanics if anyone's exploring similar prediction market structure — this stuff applies broadly beyond just BTC/ETH markets too.

Standard disclaimer: educational only, not financial advice, prediction markets carry real risk.